No, you don’t get it. The until now completely un-recursive not-self-improving training runs will automate themselves away any day now, and ML PhD’s will become worthless!!